NBER WORKING PAPER SERIES APPEARING AND DISAPPEARING DIVIDENDS: THE LINK TO CATERING INCENTIVES. Malcolm Baker Jeffrey Wurgler

Transcription

1 NBER WORKING PAPER SERIES APPEARING AND DISAPPEARING DIVIDENDS: THE LINK TO CATERING INCENTIVES Malcolm Baker Jeffrey Wurgler Working Paper NATIONAL BUREAU OF ECONOMIC RESEARCH 1050 Massachusetts Avenue Cambridge, MA September 2003 We thank Ryan Taliaferro for research assistance. We thank Raj Aggarwal, David Backus, Brian Hall, Holger Mueller, Kevin Murphy, Lasse Pedersen, Jay Ritter, and seminar participants at Columbia, Dartmouth, NYU, the NYU Salomon Center, Ohio State, Princeton, the University of Florida, and the University of Virginia for helpful comments. Baker gratefully acknowledges financial support from the Division of Research of the Harvard Business School. An earlier version appeared under the title Why are dividends disappearing? An empirical analysis. The views expressed herein are those of the authors and are not necessarily those of the National Bureau of Economic Research by Malcolm Baker and Jeffrey Wurgler. All rights reserved. Short sections of text, not to exceed two paragraphs, may be quoted without explicit permission provided that full credit, including notice, is given to the source.

2 Appearing and Disappearing Dividends: The Link to Catering Incentives Malcolm Baker and Jeffrey Wurgler NBER Working Paper No September 2003 JEL No. G35 ABSTRACT We document a close link between fluctuations in the propensity to pay dividends and catering incentives. First, we use the methodology of Fama and French (2001) to identify a total of four distinct trends in the propensity to pay dividends between 1963 and Second, we show that each of these trends lines up with a corresponding fluctuation in catering incentives: The propensity to pay increases when a proxy for the stock market dividend premium is positive and decreases when it is negative. The lone disconnect is attributable to Nixon-era controls. Malcolm Baker Harvard Business School Morgan Hall 361 Boston, MA and NBER mbaker@hbs.edu Jeffrey Wurgler New York University Stern School of Business Suite West 4 th Street New York, NY and NBER jwurgler@stern.nyu.edu

3 I. Introduction In an important paper, Fama and French (2001) document a major time-series shift in dividend policy. Between 1978 and 1999, the fraction of their Compustat sample that pays dividends fell from 67% to 21%. They trace part of this decline to a composition effect. In recent years, an increasing fraction of firms were unprofitable but had strong growth opportunities, so they would not have been expected to pay dividends. However, even after accounting for this effect, Fama and French document a large decline in the residual propensity to pay dividends. In this sense, dividends have been disappearing since In this paper, we ask whether the catering view of dividends in Baker and Wurgler (2002) sheds light on the propensity to pay dividends. That view argues that when investor demand for payers is high (low) and Modigliani-Miller style arbitrage is limited, a stock price premium (discount) may appear on payers (nonpayers), and firms on the margin may cater to the implied investor demand in an attempt to capture this dividend premium. Thus, leaving aside its allowance of a role for sentiment, the catering view can be viewed as a disequilibrium version of the clientele equilibrium view in Black and Scholes (1974). Baker and Wurgler construct proxies for the time-varying dividend premium, or catering incentive, and find that they help to explain the aggregate rate of dividend initiation and omission. We start the current analysis by applying the methodology of Fama and French to earlier data. This leads to our first main finding: There are four distinct trends in the propensity to pay dividends between 1963 and The post-1977 decline is the largest and longest, to be sure, but the three earlier fluctuations are clearly evident. While these trends are interesting in their own right, more important for us is that they essentially quadruple the degrees of freedom available for our analysis. Our second main finding is that each of these trends can be connected 1

4 to a corresponding fluctuation in a proxy for catering incentives, the stock market dividend premium variable from Baker and Wurgler. This variable, measured annually, is defined as the log difference in the value-weighted average market-to-book of payers and the value-weighted average market-to-book of nonpayers. Specifically, the dividend premium is positive in the mid-1960 s, coinciding with the first (increasing) trend in the propensity to pay that we document. Then it falls to negative territory through 1969, suggesting a premium for nonpayers, and accurately predicts the onset of the second (decreasing) trend. The dividend premium goes positive once again in 1970 and remains positive through While the propensity to pay does not begin its third (increasing) trend until 1973 or 1975, depending on how this variable is constructed, there is a simple explanation for the brief misfit. In the early 1970 s, Nixon s Committee on Interest and Dividends actively discouraged increases in dividends in an effort to fight inflation. Once their artificial controls were lifted, however, the propensity to pay immediately resumed alignment with catering incentives. Most striking of all, the dividend premium goes back to negative values in 1978 and remains negative essentially through Thus it accurately predicts both the onset and continuation of the fourth (decreasing) trend, the post-1977 disappearance. Further analysis firms up the link between catering incentives and the propensity to pay. Going beyond a qualitative correspondence, we find that the dividend premium is able to explain the actual magnitude of the post-1977 disappearance in an out-of-sample test. We also find that the dividend premium and changes in the propensity to pay forecast the relative stock returns of payers and nonpayers, which bolsters the argument that these variables were associated with a real or perceived mispricing driven by investor demand. 2

5 Finally, we conduct an exhaustive review of historical New York Times articles pertaining to dividends to better understand fluctuations in the investor demand for payers. This review suggests an intuitive pattern. The dividend premium tends to be negative, and the propensity to pay to decrease, when sentiment for growth stocks (characteristically nonpayers) is high, such as in the late-1960 s and late-1990 s. Following crashes in growth stocks, demand appears to favor the safe returns on payers, the dividend premium rises, and dividends appear. This appears to characterize the mid-1960 s, early- to mid-1970 s, and perhaps the period of this writing. In sum, our results profitably marry the work of Fama and French (2001) and Baker and Wurgler (2002). While more research on the demand side is necessary, our results establish that the catering view of the supply side offers an empirically successful account of the post-1977 disappearance of dividends as well as earlier appearances and disappearances. Of course, our results do not rule out that other influences affect the propensity to pay indeed, recent work finds some effect of repurchases (Grullon and Michaely (2002)), executive stock options (Fenn and Liang (2001)), and asymmetric information (Amihud and Li (2002)) but they raise the bar from explaining one trend to explaining four. This paper proceeds as follows. Section II documents four trends in the propensity to pay dividends between 1963 and Section III matches these to catering incentives and Nixonera controls. Section IV explores evidence on investor demand. Section V concludes. II. Four trends in the propensity to pay dividends, Here we use the methodology of Fama and French (2001) to describe the evolution of the propensity to pay dividends from 1963 through Our sample is defined as there (p ): The Compustat sample for calendar year t includes those firms with fiscal year-ends in t that 3

6 have the following data (Compustat annual data items in parentheses): total assets (6), stock price (199) and shares outstanding (25) at the end of the fiscal year, income before extraordinary items (18), interest expense (15), [cash] dividends per share by ex date (26), preferred dividends (19), and (a) preferred stock liquidating value (10), (b) preferred stock redemption value (56), or (c) preferred stock carrying value (130). Firms must also have (a) stockholder s equity (216), (b) liabilities (181), or (c) common equity (60) and preferred stock par value (130). Total assets must be available in years t and t-1. The other items must be available in t. We exclude firms with book equity below $250,000 or assets below $500,000. To ensure that firms are publicly traded, the Compustat sample includes only firms with CRSP share codes of 10 or 11, and we use only the fiscal years a firm is in the CRSP database at its fiscal year-end. We exclude utilities (SIC codes ) and financial firms (SIC codes ). The average number of firms in our sample is 1,776 between 1963 and 1977 and 3,797 between 1978 and Size, investment opportunities, profitability characteristics, and dividend payment are also defined as in Fama and French. NYP is the NYSE market capitalization percentile, i.e. the fraction of NYSE firms having equal or smaller capitalization than firm i in year t. M/B is the market-to-book ratio, defined as book assets minus book equity plus market equity all divided by book assets. Market equity is fiscal year closing price times shares outstanding. Book equity is stockholders equity (or first available of common equity plus preferred stock par value or book assets minus liabilities) minus preferred stock liquidating value (or first available of redemption value or carrying value) plus balance sheet deferred taxes and investment tax credit (35) if available and minus post retirement assets (330) if available. Growth in book assets da/a is selfexplanatory. Profitability E/A is earnings before extraordinary items plus interest expense plus 4

7 income statement deferred taxes (50) divided by book assets. A firm-year observation is a payer if it has positive dividends per share by the ex date, else it is a nonpayer. The top panels of Figure 1 show the actual percentage of the sample that pays dividends in each year, as well as the percentage of firms that would be expected to be payers given their characteristics. The expected percentage is based on firm-level logit models of the probability that a firm with given characteristics is a payer. Each year between 1963 and 1977, we estimate two models. One includes NYP, M/B, da/a, and E/A, and the other excludes M/B. (As noted by Fama and French, M/B wears several theoretical hats and so it is useful to establish robustness of various results to its exclusion.) The average yearly coefficient from these regressions, known as Fama-MacBeth estimates, imply the following models: M da E Pr ( Payerit = 1) = logit NYPit (1) B it A it A it and da E Pr ( Payerit = 1) = logit NYPit (2) A it A it The expected percentage of dividend payers in the Compustat sample in year t is then estimated by applying Eqs. (1) and (2) to the values of the explanatory variables for each firm, summing over firms, dividing by the number of firms and multiplying by The propensity to pay dividends is defined as the difference between the actual percentage and the expected and is plotted in the middle panels. The change in the propensity to pay is plotted in the bottom panels. While the exact timing of the breaks vary depending on how one measures the propensity to pay, the figure reveals four clear trends: (i) an increase from 1963 through ; (ii) a 1 The in-sample nature of the expected payer estimates is not problematic. The year-by-year model coefficients are relatively stable, so similar results obtain if the training period is instead or

8 decrease from through ; (iii) an increase from through 1977; and (iv) the decrease from 1978 onward identified by Fama and French. Each trend involves hundreds if not thousands of firms. Thus, while the latest decline has understandably received the most attention, dividends have to some extent appeared and disappeared before. III. Catering incentives and the propensity to pay dividends Here we document that the four historical trends in the propensity to pay dividends roughly coincide with four broad fluctuations in catering incentives. Once account is taken of the intervention by the Nixon administration in the early 1970 s, the correspondence is all but perfect. We then show that these forces can statistically explain the post-1977 disappearance of dividends in an out-of-sample test. We close with a complementary analysis of stock returns. A. Catering incentives Baker and Wurgler (2002) suggest that managers may try to cater to prevailing investor demand for dividends by paying dividends when investors are putting a premium on dividend payers, and not paying when the dividend premium is negative. While surely not the only omitted influence on dividend payment in Eqs. (1) and (2), catering incentives vary over time and to a large extent may be separate from firm characteristics. It is natural to examine whether they influence the propensity to pay. We measure catering incentives between 1962 and 1999 using the dividend premium variable in Baker and Wurgler. It is defined as follows. Each year, we compute the book-valueweighted average market-to-book ratio for dividend payers and the average for non-payers. The dividend premium is the difference between the logs of these averages. The market-to-book ratio 6

9 used here is defined using the calendar-year end stock price, rather than the fiscal-year end price, but otherwise follows the definition given above. Baker and Wurgler (2002) find that this variable is significantly correlated with other plausible measures of investor demand for dividends, including a high correlation with a dividend premium variable based on the dual-class structure of the Citizens Utilities company, and a significant positive correlation with the average announcement effect of dividend initiators. It is also significantly negatively correlated with the future returns of a portfolio that is long payers and short nonpayers (although a formal predictive relationship is not established there). These correlations suggest that the dividend premium variable, while quite crude, is a reasonable candidate for measuring the relative investor demand for payers. The catering theory involves dynamics in disequilibrium, and thus essentially maintains that uninformed investor demand for dividend payers fluctuates faster than firms can or do adjust. A nontrivial dividend premium (or discount) appears, and firms are presumed to cater to the implied excess demand. The appropriate comparison is thus between changes in the propensity to pay and the beginning-of-period level of the dividend premium. Figure 2 plots the (lagged) dividend premium against the annual changes in the propensity to pay. For convenience, however, we also show plots against the level of the propensity to pay. Figure 2 illustrates an impressive degree of agreement between the two series. In terms of the four trends: (i) The dividend premium predicts an increasing propensity to pay between 1963 and 1967 (that is, it takes positive values between 1962 and 1966), and the propensity to pay is indeed rising between 1963 and 1966 (no M/B) or 1968 (M/B), an essentially perfect fit. (ii) The dividend premium then goes negative to predict a declining propensity to pay from 1968 through This predicted turning point is also borne out in the data. 7

10 Regarding trend (iii), the dividend premium flips sign again in the early 1970 s, its lag predicting a rising propensity to pay from 1971 through However, there are a few years of misfit here: The propensity to pay doesn t actually start rising until 1973 (no M/B) or 1975 (M/B). After that, however, it does rise through 1977 by both measures. Thus, although the second appearance of dividends did ultimately occur, it was predicted a few years too early. (More on this period below.) Finally, the dividend premium s most striking success is in predicting the (iv) post-1977 disappearance. The dividend premium falls sharply around this period, and goes from positive to negative precisely in Moreover, it remains negative through the end of the data, except for a brief flirtation in Thus it predicts not only the onset but the continued fall in the propensity to pay over this long period. In sum, the dividend premium has run through four positive/negative cycles in this sample period, and these correspond closely with the four observed trends in the propensity to pay. Note that there is no case in which changes in the propensity to pay predate changes in the dividend premium, and only one period in which the lag was substantial. The results suggest that catering incentives may have a central effect on the propensity to pay. B. Nixon-era dividend controls From August 15 through December 31, 1971, the Nixon Administration declared a dividend freeze as part of its efforts to curb inflation. In November of 1971, the Committee on Interest and Dividends announced that corporations should observe a 4 percent dividend growth guideline in declaring dividends effective January 1, The base for this calculation was the maximum of total dividends per share paid in any of the three prior fiscal years and so a corporation that paid zero dividends per share in these years would, under the text of the 8

11 guideline, also be limited to zero dividends in Essentially similar guidelines remained in place until the Committee was dissolved on April 30, While compliance with these guidelines was ostensibly voluntary, the Administration put heavy pressure on corporations to comply with the President s request (New York Times 11/3/1971), and the available evidence indicates that the policy had bite. In the first several months of the program, the Committee monitored 7,000 firms and requested that a dividend increase be rolled back by only 29, most of which met the request (New York Times 5/7/1972). Related evidence appears in Dann (1981), who finds that repurchases, which did not violate the controls, spiked in 1973 and The Nixon controls likely explain the dividend premium s misfit in the early 1970 s. The years 1972, 1973, and 1974 are shaded in Figure 2. The controls appear to have kept the propensity to pay in decline even though catering incentives pointed the other way. Once the controls were lifted, the propensity to pay realigned with catering incentives. C. Regressions and an out-of-sample test It is clear from the plots that the dividend premium will predict changes in the propensity to pay. Table 1 confirms this formally. We report univariate regressions that include only the dividend premium and bivariate specifications that include a dummy for the Nixon controls. t D ND t PTP = a + bp 1 + cnixon + v (3) t t As discussed above, this specification is appropriate for a disequilibrium theory like catering. We run (3) for both versions of the PTP series, and for the full sample and the subperiod. The dividend premium P D-ND is standardized in this regression, so the results indicate that a one-sd higher level of catering incentives (about an 18 percentage point higher value of the 9

12 dividend premium) is associated with a one- to one-and-a-half percentage point increase in the propensity to pay dividends, while the Nixon controls appear to have reduced the propensity to pay by a few percentage points per year. Given the prominence of the post-1977 decline in the propensity to pay, an important question is whether catering incentives can empirically explain it. To give a precise answer to this question one must stay faithful to Fama and French s empirical framework. To review, they use the Compustat data to fit a model of the expected percentage of payers, and then they evaluate this model at the sample characteristics that prevail from 1978 and forward to make a true out-of-sample forecast of the expected percentage in each year. The difference between the actual and the expected percentage is the propensity to pay. By analogy, the way to determine whether catering incentives could account for the decline within this framework is to first take the propensity to pay variable as data. Then fit a regression model between the propensity to pay and the (lagged) dividend premium over the series, and use the fitted model to forecast the expected propensity to pay from 1978 forward. To the extent that the actual decline in the propensity to pay lines up with this forecast, the disappearance is explained, as an empirical matter. Table 2 shows that the dividend premium is able to account for the magnitude of the post disappearance. 3 One can calculate that the average absolute forecast error when market-tobook is included is only 3.4 percentage points (usually positive), and 4.0 percentage points (usually negative) when it is excluded. The table also reports out-of-sample forecasts made by the bivariate model that includes the Nixon dummy. This brings the average absolute forecast 2 Perhaps unaware of the controls, Dann expresses some puzzlement at the data (p. 121), but Bagwell and Shoven (1989) make the connection. 10

13 error down to only 2.2 percentage points in the case where market-to-book is included. Given that PTP is itself surely measured with at least a few percentage points of error, these forecasts seem about as accurate as one could reasonably expect. D. Evidence from returns Some additional evidence consistent with catering comes from stock return predictability regressions. Baker and Wurgler (2002) find that the aggregate rate of dividend initiation and omission predict the relative stock returns of payers and nonpayers. When initiations (omissions) are common, returns on payers are relatively low (high) over the next one to three years. The results are consistent with the joint hypothesis of catering-motivated decisions and mediumhorizon reversal of relative mispricing. 4 We briefly extend this analysis by testing the predictive power of the two variables focused on here the dividend premium variable itself and changes in PTP. We view predictive power for the dividend premium as more of an assumption of the catering view, and predictive power for changes in PTP as more of a novel implication. As the difference between two scaled-price variables, the dividend premium might be expected to have some power to predict the difference in returns of payers and nonpayers. Panel A of Table 3 generally confirms this. The table reports both OLS coefficients and coefficients adjusted for the Stambaugh (1999) small-sample bias. We use the (standardized) dividend premium to forecast the difference between the annual returns on value-weighted indexes of payers and nonpayers. We find that a one-sd increase in the dividend premium is associated with a relative return on payers over nonpayers that is 3.6 percentage points lower in the first 3 Since the dividend premium-based forecasts are generated from regressions of changes on levels in Table 1, we forecast changes in the propensity to pay, starting in 1978, and then cumulate them to estimate the expected propensity to pay from year to year. 4 One can also imagine a somewhat less stark model in which managers are not explicitly motivated by perceived stock market mispricing, but simply cater to extreme investor demands. 11

14 year ahead, 8.9 percentage points lower in the second year ahead, and 10.9 percentage points lower in the third year ahead. The results for the one and two-year ahead returns are significant using the bootstrap described in Baker and Wurgler (2002). Changes in the propensity to pay might also be expected to have some predictive power, given the aforementioned results for the raw rate of initiations and omissions. The mechanical connection between those raw rates and the perhaps more interesting changes in PTP variable is hardly exact, however, because of new lists, delists, and the netting out of firm characteristics in PTP. The effect of all these adjustments is a priori unclear, but the lower panels of Table 3 generally support the hypothesized pattern. The independent variable in each of these panels is a version of the (standardized) change in PTP. The results suggest that a one-sd increase in the propensity to pay is associated with future relative returns on payers which are lower by roughly several percent per year. However, statistical significance is sensitive to horizon, which the theory does not specify. One notable pattern is the effect of including market-to-book in the PTP definition: Results that include it are marginal at best, while those that exclude it are strong. Figure 1 provides an account for this difference. Panels C and D there show that the inclusion of market-to-book appears to add highfrequency noise to the four low-frequency trends. The fact that the predictability evidence gets stronger when this variation is cut out tends to increase our confidence that it is genuine. Overall, the results provide some additional evidence that appears consistent with the catering story. IV. Investor demand for payers and nonpayers The results above establish a tight link between a proxy for catering incentives and the propensity to pay dividends. This is the main message of the paper. Taking this as evidence that 12

15 catering motives are important to the supply of paying firms, the rest of the story pertains to the demand side. To what investor demand are firms catering? There are two broad possibilities: traditional dividend clienteles, such as those outlined by Miller and Modigliani (1961) and Black and Scholes (1974), and some notion of investor sentiment. Our first approach to understanding the demand side is to extend the Baker and Wurgler (2002) analysis of the raw rate of initiations and omissions and regress changes in PTP directly on three proxies for dividend clienteles. Because the approach is similar, we summarize the exercise only briefly. We form three clientele proxies. The personal tax advantage for dividends (in practice, a net disadvantage) is the relative after-tax income from dividends versus long-term capital gains for individual investors as calculated by the NBER TAXSIM model (Feenberg and Coutts (1993)). The corporate tax advantage of dividend income as the relative after-tax income from dividends versus long-term capital gains for C corporations corporate rates are from Graham (2003), earlier are from various issues of the IRS Statistics of Income. Finally, to proxy for transaction cost-driven clienteles, total one-way trading costs are defined as one-half the average bid-ask spread on DJIA stocks plus the average commission on round-lot NYSE transactions. The transaction cost data are from Jones (2002). We then regress these clientele proxies on changes in PTP as in Equation (3). We exclude the dividend premium from these regressions, since in the logic of the theory it represents a summary statistic for excess clientele demand. We find that none of these three proxies has a robust effect on the change in PTP. 5 Indeed, the tax clientele proxies are consistently the wrong sign. While these proxies surely mis-measure clientele demands to some extent, they appear to be straightforward measures and have been used in prior work. 5 A table is available upon request. 13

16 Our second approach to understanding demand is an exhaustive analysis of New York Times articles pertaining to dividends. We use the search engine Factiva to identify all Times articles published between January 1, 1969 (when Factiva coverage begins) and December 31, 2001 that contain dividend or dividends at least twice in the abstract. 6 This leads to an initial sample of 1,567 unique articles to inspect more closely. We read the abstract of each of these articles to determine whether the article may contain some discussion of dividends that goes beyond firm characteristics, and hence may be relevant to understanding the propensity to pay. Most articles do not satisfy this screen. Ultimately, 103 of the initial set of abstracts suggest that the article contains some useful commentary, and we read the full text of each of the 103 articles from the New York Times archives. Reassuringly, we find that a large number of these stories are suggestive of time-varying catering incentives. In particular, while references to clienteles based on transaction costs or institutional investment constraints are almost nonexistent, there are many mentions of dividends in the context of taxes and investor sentiment. However, many of the tax references involve tax policy proposals that were ultimately not implemented. For instance, those that appear around the crucial turning point in PTP include a proposal to eliminate double taxation of dividends and a proposal to withhold taxes on dividend income. Both proposals were defeated and thus could have no persistent effect on PTP. More generally, there is little agreement between the timing and content of the tax-related articles and actual fluctuations in the dividend premium or PTP. This seems consistent with our own non-results described above. The sentiment references provide more affirmative evidence. While Baker and Wurgler (2002) report a correlation between the dividend premium and the closed-end fund discount, 6 The search engine for historical Wall Street Journal articles is not sufficiently precise. For example, one cannot exclude advertisements. This causes thousands of false hits and makes the analysis unmanageable. 14

17 suggestive of some role for sentiment, the news reports give more descriptive color. Table 4 summarizes the basic pattern through a small sample of Times quotations. Which ones to present is somewhat arbitrary; these are chosen because they span a wide period and capture themes that appear repeatedly. 7 The interesting thing to note here is that the clusters of sentiment-related references coincide with observed patterns in the dividend premium and the propensity to pay. For instance, almost all of the references to sentiment for dividends appear in the years just before 1977, as the dividend premium was high and the propensity to pay was rising to a local maximum. The references that suggest sentiment was against payers occur after that, when the propensity to pay was low and falling. Some of these quotations also make reference to the late-1960 s market environment (which predates Factiva coverage). They describe a boom in sentiment for new issues and extreme-growth stocks that is similar to the late-1990 s boom. This parallel is also noted by Malkiel (1999). In both of these eras, of course, the data suggest a discount on payers and a declining propensity to pay. While there are clear limits to this sort of analysis, it sheds some useful, if preliminary, light on the nature of the demand for payers. In particular, Table 4 (and the articles that could not be included there) suggests that the patterns we observe are affected by booms and busts in extreme-growth stocks, characteristically nonpayers. When sentiment for growth is high, as in the late 1960 s and the late 1990 s, the marginal investor appears to demand opportunities for capital gains, not staid firms that pay dividends. The dividend premium is negative and dividends tend to disappear. When this sentiment reverses, typically following a crash, the marginal investor demands stocks with perceived safe features including dividends. This seems to 7 Complete details of the database search procedure, classification, and results are available upon request. 15

18 characterize the mid-1960 s (which followed an early-1960 s crash in growth stocks) and the mid-1970 s, where the dividend premium rises and the propensity to pay increases. 8 V. Conclusion We establish a close empirical link between the propensity to pay dividends and catering incentives. First, we apply the methodology of Fama and French (2001) to earlier data to identify four distinct trends in the propensity to pay between 1963 and 2000 two appearances and two disappearances. Second, we show that each of these trends is associated with a corresponding fluctuation in catering incentives, where the latter is measured by the dividend premium variable from Baker and Wurgler (2002). Once the impact of the early 1970 s intervention by Nixon s Committee on Interest and Dividends is noted, our analysis addresses essentially all significant fluctuations in the post-1963 propensity to pay time series. Moreover, we show that catering incentives are able to explain, in the appropriate out-of-sample test, the actual magnitude of the post-1977 disappearance documented by Fama and French. A review of historical articles from the financial press suggests that firms may be catering to sentiment-driven demand. Dividends tend to disappear during pronounced booms in growth stocks, and reappear after crashes in such stocks. The next few years will offer an out-of-sample test of this dynamic: Internet stocks have recently crashed and market observers characterize the current period as a bear market. If these market conditions continue, history suggests that the dividend premium will rise and dividends will reappear. 8 Consistent with this dynamic, Fuller and Goldstein (2002) find that payers have higher (less negative) returns than nonpayers in months in which the S&P Index return is negative. This holds after controlling for factor loadings. 16

19 References Amihud, Yakov, and Kefei Li, 2002, The declining information content of dividends and the effect of institutional holdings, New York University working paper. Bagwell, Laurie Simon, and John B. Shoven, 1989, Cash distributions to shareholders, Journal of Economic Perspectives 3, Baker, Malcolm, and Jeffrey Wurgler, 2002, A catering theory of dividends, Journal of Finance, forthcoming. Black, Fischer, and Myron S. Scholes, 1974, The effects of dividend yield and dividend policy on common stock prices and returns, Journal of Financial Economics 1, Dann, Larry Y., 1981, Common stock repurchases: An analysis of returns to bondholders and stockholders, Journal of Financial Economics 9, Fama, Eugene F., and Kenneth R. French, 2001, Disappearing dividends: Changing firm characteristics or lower propensity to pay?, Journal of Financial Economics 60, Feenberg, Daniel, and Elizabeth Coutts, 1993, An introduction to the Taxsim model, Journal of Policy Analysis and Management 12, Fenn, George W., and Nellie Liang, 2001, Corporate payout policy and managerial incentives, Journal of Financial Economics 60, Fuller, Kathleen, and Michael Goldstein, 2002, The ups and downs of dividend preference, University of Michigan working paper. Graham, John R., 2003, Taxes and corporate finance: A review, Review of Financial Studies, forthcoming. Grullon, Gustavo, and Roni Michaely, 2002, Dividends, share repurchases, and the substitution hypothesis, Journal of Finance 57, Jones, Charles, 2002, A century of stock market liquidity and trading costs, Columbia University working paper. Malkiel, Burton G., 1999, A Random Walk Down Wall Street, (Norton, New York, NY). Miller, Merton H., and Franco Modigliani, 1961, Dividend policy, growth and the valuation of shares, Journal of Business 34,

20 Figure 1. The propensity to pay dividends, Panels A and B show the actual percent (solid) and expected percent (dashed) of dividend payers in Compustat. Panels C and D show the propensity to pay dividends, i.e. the difference between the actual and expected percent. Panels E and F show changes in the propensity to pay dividends. Actual percent is the number of dividend payers divided by the number of firms in the sample that year. Expected percent is the expected percent of dividend payers based on prevailing sample characteristics. Following Fama and French (2001), one set of results (Panels A, C, and E) estimates the expected percent of payers with a logit model that includes the NYSE market capitalization percentile, the market-to-book ratio, asset growth, and profitability. The other set (Panels B, D, and F) excludes market-to-book. The propensity to pay dividends PTP is the difference between the actual and expected percent. Panel A. Actual and expected percent payers Panel B. Actual and expected percent payers (no M/B) Panel C. The propensity to pay dividends Panel D. The propensity to pay dividends (no M/B) Panel E. Changes in the propensity to pay Panel F. Changes in the propensity to pay (no M/B)

21 Figure 2. The dividend premium and changes in the propensity to pay. Panel A shows the value-weighted dividend premium from Baker and Wurgler (2002) in percentage terms (lagged once; dashed line; right axis), changes in the propensity to pay dividends (solid line; left axis), and shading to denote the 1972 through 1974 Nixon dividend controls era. Panel B shows changes in the propensity to pay estimated from a measure that excludes market-to-book as a firm characteristic. Panel A. The dividend premium, Nixon controls, and changes in the propensity to pay Panel B. The dividend premium, Nixon controls, and changes in the propensity to pay (no M/B)

22 Table 1. Regressions to explain changes in the propensity to pay. Changes in the propensity to pay dividends regressed on the lagged value-weighted dividend premium and a dummy for the 1972 through 1974 Nixon dividend controls period: D ND PTP t = a + bpt + cnixont + v. 1 t The dividend premium P D-ND is standardized to have unit variance. The left panels report regressions where PTP is estimated using a firm-level model of dividend payment that includes market-to-book as a relevant firm characteristic. The right panels report regressions where market-to-book is not included. T-statistics use standard errors that are robust to heteroskedasticity and serial correlation up to four lags. M/B t included M/B t excluded b [t] c [t] b [t] c [t] Panel A Univariate 1.04 [2.4] [3.2].. Bivariate 1.53 [4.8] [-4.7] 1.22 [3.1] [-1.3] Panel B Univariate 1.19 [1.7] [5.7].. Bivariate 1.70 [2.2] [-3.9] 1.34 [5.0] [-1.4]

23 Table 2. Out of sample forecast of the propensity to pay dividends. Actual percent is payers divided by firms. To determine the expected present, we run Fama-MacBeth logit regressions of dividend policy on firm characteristics, using firm-year observations from 1963 to The firm characteristics are the NYSE percentile NYP, asset growth da/a, and profitability E/A. The six four columns also include the market-to-book ratio M/B, while the second six columns do not. Expected percent of payers for a year t is estimated by applying the average logit regression coefficients for to the values of the characteristics for each firm for year t, summing over firms, dividing by the number of firms and then multiplying by 100. The propensity to pay PTP is the actual percent minus the expected percent. Expected PTP are the forecast values from the second stage regression presented in Table 1. Base data Dividend Premium Actual Expected Percent Expected PTP PTP M/Bt included M/Bt excluded PTP Expected PTP Dividend Premium, Nixon Base data Dividend Premium Expected PTP PTP Expected Expected PTP Expected PTP PTP PTP Expected PTP Dividend Premium, Nixon Expected PTP PTP Expected PTP

24 Table 3. Changes in the propensity to pay dividends: Predicting returns, Univariate regressions of future excess returns of dividend payers and nonpayers on the changes in the propensity to pay dividends. The dependent variable in is the difference in future returns between dividend payers and nonpayers. r t+k denotes returns in year t+k, and R t+k denotes cumulative returns from t+1 through t+k. In panel A, the independent variable in panel is the value-weighted dividend premium from Baker and Wurgler (2002) in percentage terms. In Panels B through E, the independent variable is the change in the propensity to pay dividends. In Panels B and D, the propensity to pay is estimated including the market-to-book ratio as a firm characteristic. In Panels C and E, the propensity to pay is estimated without the market-to-book ratio. Panels D and E make an adjustment for the influence of the Nixon controls, using the Nixon coefficients from Table 1. All independent variables are standardized to have zero mean and unit variance. We report OLS coefficients and bias-adjusted (BA) coefficients. Bootstrap p-values represent a two-tailed test of the null of no predictability. N OLS BA [p-val] R 2 Panel A. Dividend premium r Dt+1 r NDt [0.25] 0.07 r Dt+2 r NDt [0.03] 0.17 r Dt+3 r NDt [0.02] 0.22 R Dt+3 R NDt [0.07] 0.33 Panel B. Changes in PTP (M/B included) r Dt+1 r NDt [0.90] 0.00 r Dt+2 r NDt [0.08] 0.08 r Dt+3 r NDt [0.15] 0.06 R Dt+3 R NDt [0.32] 0.10 Panel C. Changes in PTP (M/B excluded) r Dt+1 r NDt [0.19] 0.05 r Dt+2 r NDt [0.00] 0.34 r Dt+3 r NDt [0.02] 0.22 R Dt+3 R NDt [0.02] 0.47 Panel D. Changes in PTP (M/B included), Nixon adjustment r Dt+1 r NDt [0.96] 0.00 r Dt+2 r NDt [0.05] 0.10 r Dt+3 r NDt [0.05] 0.12 R Dt+3 R NDt [0.23] 0.14 Panel E. Changes in PTP (M/B excluded), Nixon adjustment r Dt+1 r NDt [0.22] 0.05 r Dt+2 r NDt [0.00] 0.33 r Dt+3 r NDt [0.01] 0.23 R Dt+3 R NDt [0.04] 0.47

25 Table 4. New York Times articles. Selected articles from a Factiva search of New York Times articles over the period from 1969 through Article date Sentiment for dividends Quotation 11/7/1976 Yes Thanks to [characteristics and] the rising yield-consciousness of stockholders, corporations are fattening their dividend payouts. As investors became chary of the stock market, they were less apt to count on future earnings growth and more likely to return to the bird-in-the-hand rationale of cash dividends. 5/18/1977 Yes After years of disappointment particularly with low-yielding glamour stocks investors are emphasizing dividends in their stock selections. 2/15/1981 No nondividend payers have become the Big Board s star performers in recent years My sophisticated investors never ask me if a stock pays a cash dividend, says Mr. Schaeffer of Bache. They d much rather have stock dividends than cash dividends. 5/7/1995 No These days, dividends are rising rapidly, but not as fast as stock prices perhaps we are witnessing a sea change in investor attitudes. Most investors don t seem to be very interested in dividends just now. Maybe dividends simply don t matter anymore. 1/3/1997 No In this buoyant stock market, companies have seen relatively little demand for higher payouts from shareholders who, after all, have been seeking and getting capital gains. 10/7/1999 No No (late 1960 s) Yes (after 1968) 1/4/2000 No No (late 1960 s) What is unusual is that the economy is doing so well even while companies are growing more reluctant to raise their dividends the [last] time companies cut back on dividend increases even as the economy continued to grow is the late-1960 s market bears more than a passing resemblance to this one. The stock market had been going up steadily for the better part of two decades Dividends can go so low because investors do not care much about them. It is capital gains that have made them rich, and it is the pursuit of capital gains that drives stock investments now. After 1968, as it became clear that capital losses were possible, investors came to value dividends, and the pressure grew on companies to pay them. A growing portion of corporate America appears to be concluding that dividends are no longer needed to attract investors decline [in percent of payers in S&P] also reflects an investor attitude that puts little pressure on companies to make payouts. The only similar trend occurred in the late 1960 s, another time that small technology companies were all the rage and the market for new issues was red hot. A variety of reasons are given for the trend away from dividends, including the tax disadvantages but that has always been true, and the effect presumably should have been greater two decades ago, when tax rates were much higher The most likely explanation would seem to be the most obvious. Investors, after seeing year after year of huge capital gains, no longer see much of a need for dividends as an assured return if the market declines